Put transaction log files on a separate physical disk from the database.

This is the single most important configuration affecting the performance of Exchange. This configuration also has recovery implications, since transaction logs provide an additional recovery resource.

Make Write Cache unavailable on the SCSI controller.

The Windows operating system does not use buffers, so when Exchange receives a write complete notice from Windows, the write-to-disk has been completed. If Write Cache is enabled, Windows responds as though a write-to-disk has been completed, and will provide this information to Exchange (or other applications) incorrectly. The result could be data corruption if there is a system crash before the operation is actually written to disk.

Make circular logging unavailable if possible.

Circular logging minimizes the risk that the hard disk will be filled with transaction log files. But, if a solid backup strategy is in place, transaction log files are purged during the backup, thus freeing disk space. If circular logging is enabled, transaction log histories are overwritten, incremental and differential backups of storage groups and databases are disabled, and recovery is only possible up to the point of the last full or copy backup.

Avoid making the Exchange Server a domain controller.

For disaster recovery purposes, it is much easier to restore Exchange if you don't have to restore the Active Directory first.

Install Exchange into a domain that has at least two domain controllers.

Active Directory replication is not possible with only one domain controller in a domain. If the domain controller fails and corrupts the Active Directory, some transactions may not be recoverable if they were not included with the last backup. With at least two domain controllers in a domain, databases on the failed domain controller can be updated using replication to fill in missing transactions after the database backups have been restored.